Parallel and explicit methods for the simulation of eddy current problems

点击翻译按钮获取中文摘要

This research project Parallel and Explicit Time Integration Methods for the Simulation of Eddy Current Problems continues the work started in a forerunner project funded by the DFG grant. The objective of this project is to develop efficient parallel algorithms for the simulation of eddy current problems that are well suited for current heterogeneous massively parallel computer architectures. Since the parallelization of conventional implicit and the recently developed (semi-)explicit time integration methods for magneto-quasistatic problems might not scale well beyond a specific number of cores, also the application of parallel-in-time methods like Multiple Shooting or Parareal is proposed. These methods show new additional potential for an efficient parallelization along the time axes.The newly developed algorithms aim to allow for a faster computation of magnetic fields e.g. in three-dimensional high-resolution models of transformers, actuators and electric machines than established methods. This enables optimization and uncertainty quantification and is especially relevant e.g. in the design process of new inverter-fed electric machines for electric vehicles. Within the previously funded research projects the development of parallel and (semi-)explicit time integration methods set the foundations for the further acceleration of the magnetic field simulations and an even extended use of GPU acceleration. Within this following-up research project, these recently developed algorithms are to be improved and further acceleration of high-resolution three-dimensional magnetic field simulations models is to be achieved by additionally adopting and extending the Parareal methodology.